The 2.5D method stacks multiple adjacent slices into an input similar to an RGB three-channel format; the most common implementation takes the target slice and its adjacent slices immediately before and after it and places them into the RGB three channels, allowing pretrained 2D networks to be directly reused.